NVIDIA process & pain-point quotes — sourcing for the simulator
What this is. For each step in the simulator’s process diagram (and each headline pain point), the most specific direct quote from the three NVIDIA conversations — word-for-word — with speaker and source. Where no quote actually articulates a step, it’s marked NO DIRECT QUOTE so we don’t put words in their mouth.
Read this first:
- Quotes are raw auto-transcription (Granola). They contain artifacts (“RNA” = RMA, “AS and” = ASN, “SOA” = SLA, “past you on” = past due on, “scout” = scale). I’ve left them verbatim and added
[clarification]only where an artifact obscures meaning. Light cleanup is fine before display — flag if you want me to produce cleaned versions. - Speaker attribution: the Granola transcript labels the whole NVIDIA side as one voice, so on the Greg + Lonny call I’ve attributed by content where confident, else “Greg/Lonny.”
- The Lonny phone chat is Dustin recapping a call, so those are second-hand — usable as themes, not as on-record quotes.
Part 1 — Process steps (simulator walkthrough order)
1. Customers (installed fleet — where RMAs originate)
“now that we have like the scaled Operations so big. Right? We can’t simply process the amount of returns and everything from all the CSPs.” — Alex Zhu, May 27. (CSP = cloud service provider; he names “meta google core wave XAI… open[AI]”.)
Supporting (second-hand): “we have customers… they’re like so far in our back pocket right now. Not because they want to, but because they have to.” — Lonny, relayed by Dustin, Jun 17.
2. Production → Sale of new GPUs
“the company has been really prioritizing and favoring the mass production versus the RMA so we don’t really have a good RMA process” — Alex Zhu, May 27.
3. Entitlement triage — “whose problem is it, ours or the customer’s?”
NO DIRECT QUOTE that frames a fault/ownership triage (NVIDIA-vs-hyperscaler) as a step. This split came from our own modeling, not from them. Closest real quote is warranty disposition (a different kind of triage):
“we have the standard warranty as well as extend the warranty… the standard warranty is the you know the lead time or the SLAs will will be a lot higher than… extended warranties so the human being have to basically be able to disposition and then treat them so differently” — Alex Zhu, May 27.
4. Reject — not ours → back to customer
NO DIRECT QUOTE. Neither call describes bouncing out-of-warranty / customer-caused units back to the customer. (This is an inference in our model.)
5. Repair triage — repair vs. replace
“say they return 100 and we’re able to repair so what do we do for the remaining 40? So we have to we have to skew this 40 from vertical steel this already from new buy” — Alex Zhu, May 27. (i.e., ~60 of 100 get repaired; the rest are filled from new inventory.)
Supporting:
“the chip is rarely the problem. It’s usually something else on that assembly that is the problem… we salvage a very, very high percentage of our, our chips.” — Greg, Jun 17.
6. Owed-back / reconciliation (cores customers owe back; no visibility)
“we’re really good at pushing this product out to them. But then everybody kind of walks away… I, well, hold on a minute. Like, where’s what they owe us?” — Lonny, Jun 17.
“if they come here and they see that they’re past you on 155 units, they’re probably going to be less likely to ask us for more.” — Greg, Jun 17.
[past you = past due]— this is the real “155” anchor.
7. Contract manufacturers (offsite repair — Wistron/Foxconn)
“we have to move the stuff from the warehouse to the C[M]… we just give the broken parts to the repair facility say hey by the way fix this fix that so the repair facility they have the suspect sheet think about it super early 90s” — Alex Zhu, May 27.
“we’re not, I don’t, to my knowledge, we’re not repair[ing] gpus… Mostly you get into the network switches. There’s things on the network switches that you can repair.” — Greg, Jun 17. (Confirms repairs are board-level, done by CMs — NVIDIA doesn’t repair the GPU itself.)
8. Scrap (can’t be saved)
NO DIRECT QUOTE that names scrapping a unit. The closest is the inverse — Greg saying they salvage almost everything:
“we salvage a very, very high percentage of our, our chips.” — Greg, Jun 17. (Implies a small scrap remainder, but scrap is never stated. Consider softening this node or labeling it as our inference.)
9. Service pool (refurb pool → advance / like-for-like replacement)
“What we would do in my former lives is we would repair it, and then I would put it into a services pool… We’ll bring that one in. We’ll repair it. That doesn’t mean they’re going to get the same one. You’re going to get a like for like product.” — Greg, Jun 17. (Note: “in my former lives” — he’s describing the model from a prior company as the target state.)
10. Pool runs dry → pull a new GPU off the line (lost sale)
“new by[uy] is all what jensen cares about because new buy is basically revenue you’re selling more to the customer but if we want to keep the customer happy we have to seal[steal] a portion from you to keep them happy” — Alex Zhu, May 27. (This is the exact opportunity-cost / “lost sale” mechanic — pulling from new-buy to satisfy returns.)
Part 2 — Pain points
No visibility / “the score of the game”
“we’re not really good at measuring our commitments or even understanding what our commitments are and how well we’re doing against them… So we don’t even really know the score of the game right now.” — Greg, Jun 17.
Single source of truth / siloed data
“The biggest pain point is… we don’t have a unified platform… today like all of these are in silos… single source of truth… and then we have to have a unified platform” — Alex Zhu, May 27.
“for being… one of, if not the most successful companies in the world, the data is very siloed and disparate in all these different places.” — Greg, Jun 17.
Email + spreadsheets / diverging data
“It’s a lot of email and spreadsheets. That’s what we’re trying to fix.” — Greg, Jun 17.
“it’s easier for people to just reply… Next thing you know, you’ve got 16 people on here… And now where, where was the source of truth?” — Lonny, Jun 17.
“when people have two different sets of data and they’re looking at two different sheets of music. It… ends the conversation… it’s like a non-starter.” / “we come to the conversation, and we’re not even armed.” — Greg / Lonny, Jun 17.
SLA (30 days)
“we have an S[L]A that we believe[] [is] about 30 days from when a standard RMA arrives in our warehouse to when we need to give them a replacement unit.” — Greg, Jun 17.
No escalation triggers (day 31)
“What are the escalation triggers? Right? Why is there no… if I send something to you as a customer and I expect that… day 30, that module is on its way back to me? And if it’s not, by day 31, I should start firing notifications and escalations to people.” — Lonny, Jun 17.
Macro scorecard they want
“It’s at a macro level kind of overall health of, like, how many RMAs have they submitted, how many have they closed? What’s the average turnaround time?” — Greg, Jun 17.
Repair capacity insufficient
“the amount of repair that we can do… [is] not even sufficient to cover the return.” — Alex Zhu, May 27.
Coordination failure (the trucks)
“We show up with five trucks. And I go, we can’t take them… now all those trucks need to turn around. And go back to a security yard… with hundreds of millions of dollars of equipment… Just soaking in the sun in phoenix.” — Greg, Jun 17.
Telemetry is not available
“You’re not gonna get Telemetry off of their devices… access to any of that is just, it’s never going to happen.” / “even when those units get pulled out… they’re basically… scraped [wiped].” — Greg, Jun 17.
Consignment vs. turnkey reconciliation
“consignment stands for we put the inventory in the vendor[‘s] physical location[;] they would take whatever they need… there has to be a material manager doing the reconciliation.” — Alex Zhu, May 27.
Hyperscalers escalate hard
“these hyperscalers, they’re different animals… They are not shy to pound for what they want… we need to not be shy about what we need as well.” — Lonny, Jun 17.
The fundamental problem (their own framing, confirmed)
“the fundamental problem is how you understand, communicate with and service customers.” — “Yes.” — Bliss framing, confirmed by Greg/Lonny, Jun 17.
”Target-rich environment” (their phrase)
“we’re such a target rich environment” / “we’re a Target rich environment” — Greg & Lonny (both used it), Jun 17.
Part 3 — Gaps (steps with NO direct quote)
| Simulator step | Status |
|---|---|
| Entitlement triage (ours vs. customer fault) | No direct quote — our model’s framing; closest is Alex on warranty-tier disposition |
| Reject → back to customer | No direct quote — inference |
| Scrap | No direct quote — only the inverse (“salvage a very high percentage”) |
Everything else maps to a real quote above. Recommend either (a) sourcing these three on the next call, or (b) labeling them in the walkthrough as our inference rather than attributing them to NVIDIA.
Sources: Alex Zhu, May 27; Greg + Lonny, Jun 17; Lonny via Dustin, Jun 17.